PROMPT GUIDE
How to Use LoRA (Syntax, Strength and Trigger Words)
A LoRA is a small add-on file for Stable Diffusion that teaches extra characters, art styles, outfits or poses. You can use one just by writing it in the prompt, but if the strength, trigger word or base model do not match, it may not work or may break the picture. This guide covers the usage that applies to any LoRA.
Last updated: approx. 4 min read
The short answer
In AUTOMATIC1111 or Forge, use a LoRA by writing <lora:filename:0.8> in the prompt. A strength of about 0.6 to 0.8 is typical, and adding the trigger word makes it work properly. If it has no effect, first check that the base model family (SD1.5, SDXL, Pony and so on) matches.
What is a LoRA?
A LoRA is a small add-on file that adds a character, art style, outfit and so on to a base model.
A LoRA adds a specific look or style without retraining the original model. The files are small and loaded only when used, so they are popular for keeping a character consistent, fixing an art style or drawing a specific outfit.
How do you write a LoRA in the prompt?
In AUTOMATIC1111 or Forge, write it like <lora:filename:0.8>.
In interfaces such as AUTOMATIC1111 or Forge, writing <lora:filename:strength> inside the prompt loads the LoRA. The file name is the name without the extension (.safetensors). In ComfyUI you do not write it in the prompt; instead you connect a node that loads the LoRA.
<lora:[LoRA file name]:0.8>, [trigger word], 1girl, solo, smile, upper body, masterpiece, best qualityRinka's tips
- The <lora:...> part can go anywhere in the prompt, but grouping it at the start or end is easier to read
- Put the file in the right folder (models/Lora in many interfaces). Choosing from the LoRA list in the interface avoids typos
How strong should a LoRA be?
About 0.6 to 0.8 is the most common range; follow the recommendation on the download page if there is one.
Strength is given as a number from about 0 to 1. Closer to 1 works more strongly, but too strong can break the picture or make other tags stop working. Download pages often state a recommended strength, so start there.
What is a trigger word, and where does it go?
A trigger word is the cue that calls up the LoRA; put it early, near the number-of-people tag.
A trigger word is the cue word used when the LoRA was trained. Many LoRAs work only weakly with just <lora:...>, and show up properly only when the trigger word is in the prompt too. It is usually written on the download page or in the LoRA information panel of the interface, so always check it.
Rinka's tips
- Trigger words work better near the front, close to the number-of-people tag (1girl)
- For character LoRAs, adding hair color, eye color and outfit tags along with the trigger word improves the likeness
How do you use a LoRA? (Steps)
Check the supported model, recommended strength and trigger word, write them in the prompt, and compare with and without the LoRA.
- On the download page, check the supported base model (SD1.5, SDXL, Pony, Illustrious and so on), the recommended strength and the trigger word
- Put the file in the right folder and refresh the list in the interface
- Add <lora:filename:recommended strength> and the trigger word to the prompt
- With a fixed seed, compare results with and without the LoRA to confirm it works
- If it breaks, lower the strength by 0.1 at a time; if it is weak, move the trigger word toward the front
How do you use several LoRAs at once?
Use up to 2 or 3 with different roles, and lower each strength a little.
You can use many LoRAs at the same time, but the more you add, the more they pull against each other and break the picture. Keep it to 2 or 3 with different roles, such as a character LoRA and a style LoRA, and lower each strength a little so the total does not get too high.
<lora:[character LoRA]:0.8>, <lora:[style LoRA]:0.5>, [trigger word], 1girl, solo, smile, upper bodyWhat if it does not work? (Common mistakes and fixes)
Here are common problems with their causes and fixes. Change only one thing at a time and compare with the same seed, which makes the cause easier to find.
Things to keep in mind when using LoRAs
Each LoRA creator sets their own terms, such as commercial use, publishing generated images and redistribution. Always read the terms on the download page before use, and avoid LoRAs that imitate real people or that need permission from a rights holder.
For the (tag:1.2) weight syntax: The Complete Guide to Prompt Weight Syntax and BREAK →
For prompts on Pony models: How to Write Pony Diffusion Prompts (score_9 and source_anime) →
For the tag order on Illustrious and NoobAI: Illustrious and NoobAI Prompt Order →
For character sheets that keep the same character consistent: Full-Body Standing Art, Turnaround and Character Sheet Prompts (Plus Expression Variants) →
FAQ
Frequently Asked Questions
How do I write a LoRA in the prompt?
In AUTOMATIC1111, Forge and similar interfaces, write it like <lora:filename:0.8>. The file name is the name without the extension. In ComfyUI, connect a node that loads the LoRA.
What strength should a LoRA use?
About 0.6 to 0.8 is the most common range. If the download page recommends a strength, use it, and lower it by 0.1 at a time if the picture breaks.
Why does my LoRA have no effect?
The most common reason is a base model family mismatch. Next, check that the trigger word is included and the strength is not too low.
How many LoRAs can I use at once?
There is no fixed limit, but keeping to 2 or 3 makes breakage less likely. When you add more, lower each strength a little.
Where do I find the trigger word?
It is usually on the download page or in the LoRA information panel of the interface. It may also be stored in the training information inside the file.
RINKA WORLD
Play in Rinka World
MORE
